GPT Image 2 OpenAI vs Seedream 4.0 ByteDance

GPT Image 2

• + Excellent high-resolution textures on the book leaf and wooden table
• + Accurate glass refraction and reflections including the sphere's reflection on the bottom glass
• + Strong adherence to the spatial requirements of the prompt

• − The plant is more behind the cube than showing through it compared to Model B

Seedream 4.0

• + The blue sphere has a realistic translucent glass quality
• + The plant is clearly visible through the glass as requested by the prompt
• + Captures the soft window light effect with warm highlights on the table

• − The glass cube geometry is slightly warped on the right side
• − Lower resolution for textures like the book cover and leaves

GPT Image 2

• + Excellent skin texture and facial detail with high realism.
• + Very effective shallow depth of field and authentic bokeh.
• + Accurate street scene details including Japanese text and realistic tools.

• − The bike frame and wheel structure have some physical inconsistencies near the hands.
• − The motion blur on the background car is slightly static compared to the prompt's intent.

Seedream 4.0

• + Successfully captured motion blur on the passing vehicle.
• + Stronger emphasis on the 'red bicycle' through center-weighted composition.
• + Good reflection on the wet pavement.

• − Hand anatomy is very poor with merged and mushy fingers.
• − The image quality is softer with less convincing skin texture.
• − The rain is rendered as generic vertical streaks that look artificial.

GPT Image 2

• + Exquisite engraving and weathering textures on the plate armor.
• + Complex, realistic hair braiding with integrated beads.
• + Naturalistic skin textures including subtle dirt and lifelike eyes.

• − Lighting is somewhat diffuse rather than distinct warm torchlight.
• − Background is slightly generic despite the bokeh.

Seedream 4.0

• + Strong adherence to 'warm torchlight' and 'bokeh sparks' with high contrast lighting.
• + Excellent rendering of leather straps and buckles over the armor.
• + Clearer representation of facial scars and Battle-worn appearance.

• − The hair braids appear somewhat thick and less integrated into the scalp.
• − Armour engraving is less intricate than the other model.

GPT Image 2

• + Excellent text rendering with no spelling errors across many items.
• + Professional and realistic menu layout that follows all prompt categories.
• + High-quality food photography that is consistent and appetizing.

• − The layout is very dense, pushing the limits of 'minimalist' but remaining professional.

Seedream 4.0

• + Follows a simple grid structure with bold sans-serif headers.
• + Uses high-contrast vibrant accent colors as requested.

• − Total lack of menu content such as dish names, descriptions, or prices.
• − The grid alignment is awkward and leaves too much empty whitespace.
• − Image variety is low, showing a pizza in the 'Mains' section which is redundant.

GPT Image 2

• + Perfect adherence to all text requirements including the specific €6.99 price.
• + High-quality, photorealistic texture on the meat, bun, and vegetables.
• + Excellent use of the 'fiery glowing effect' across all design elements.

• − The composition is a bit crowded with text and visual elements competing for space.

Seedream 4.0

• + Good sense of dynamic motion with the swirling light trails.
• + Creative interpretation of the 'exploded' burger showing multiple layers.

• − Failed the pricing prompt, displaying €5.99 instead of €6.99.
• − The food textures appear more oily and less realistic compared to the other model.
• − The background and text integration feel less cohesive and a bit 'pasted on'.

GPT Image 2

• + Perfect adherence to text content, including the full product names and the specific date.
• + Very realistic chalk texture with granular details and pressure sensitivity simulation.
• + Excellent composition that feels like a real café wall installation.

• − The lettering style is a bit uniform and thin for a decorative chalkboard.
• − The wooden frame is slightly cropped at the top and bottom.

Seedream 4.0

• + Bold, artistic handwriting style that fits the 'elegant cursive' and 'handwritten' prompt well.
• + Dynamic chalkboard texture with smudge marks and chalk dust at the base.
• + Good use of bolding and scale for visual hierarchy.

• − Failed to include the full text for the last item, truncating 'Brown Butter Chocolate Chip Cookies' compared to Image A.
• − The perspective is slightly angled, making parts of the board harder to read.
• − Some letters in the smaller text at the bottom become slightly distorted.

GPT Image 2

• + Perfect adherence to the pose and background from Image 1.
• + Excellent character preservation of the man's face, sunglasses, and clothing from Image 2.
• + High-quality rendering with clean anatomy and logical scarf placement.

• − The fingernails on the right hand are painted red, which was a detail from the woman in Image 1 not the character in Image 2.

Seedream 4.0

• + Successful integration of the scarf and black clothing from Image 2.
• + Maintains the general background and prop from Image 1.

• − Fails significantly on character preservation, merging the man's face with the woman's long hair.
• − The anatomical structure of the legs and the way they meet the ottoman is distorted.
• − The face has significant artifacts and lacks clear features.

GPT Image 2

• + Excellent adherence to the specific prompt requirement of putting the horse on top of the astronaut.
• + High tactile detail in the textures of the lunar surface and the astronaut's suit.
• + Creative interpretation of the harness and saddle to fit the surreal scene.

• − The leather straps have confusing intersections with the astronaut's body.
• − The horse's legs end abruptly in cup-like stirrups that lack anatomical logic.

Seedream 4.0

• + Beautiful cinematic lighting and vibrant nebular background.
• + High-quality rendering of the horse's fur and the reflective visor.

• − Completely failed the negative constraint/specific instruction to put the horse on top.
• − Standard cliché interpretation of an astronaut riding a horse.

GPT Image 2

• + Excellent preservation of the subject's exact face, expression, and skin textures
• + High-fidelity recreation of the plaid scarf pattern and pea coat texture
• + Logical hand placement that respects the coat pockets

• − Crop is slightly different from the source image, losing some of the top wood structure
• − Missed the sunglasses and shoes from the target outfit

Seedream 4.0

• + Successfully included additional accessories like the sunglasses and unique shoes
• + Captured the full-body pose including the raised leg aspect of the target image

• − Failed significantly to preserve the source person's face, adding facial hair and changing features
• − Distorted hand anatomy on the right side
• − The scarf pattern is less accurate to Image 2 than the competitor's

GPT Image 2

• + Excellent photorealistic texture on the capybara's fur and the jacket fabric.
• + The lighting transition from the cool street lights to the warm taxi interior is very cinematic.
• + Includes a more authentic-looking vintage taxi driver cap.

• − The capybara's hands look slightly more like paws with fingers rather than true capybara anatomy.
• − The perspective makes the capybara appear extremely large relative to the car.

Seedream 4.0

• + Strong composition showing more of the taxi's exterior and interior simultaneously.
• + Captures the 'bored' expression of the passenger very effectively.
• + Anatomy of the capybara's paws on the wheel is quite clear.

• − The capybara's head has a slightly less realistic transition into the neck/jacket compared to Model A.
• − The 'TAT' text on the hat is nonsensical.

GPT Image 2

• + Excellent typography with perfect spelling in all requested text fields.
• + Highly detailed and intricate gothic border containing spiders, thorns, and architectural elements.
• + Strong composition that naturally integrates NYC architecture with spooky elements.

• − The parchment texture is a bit dark in the bottom half, making the small text slightly harder to read.

Seedream 4.0

• + Dynamic lighting on the jack-o-lantern and surrounding environment.
• + Successful incorporation of all requested elements including thorns and webs.

• − Spelling errors in the scroll banner text ('invicad', 'frizhis').
• − The border feels less integrated and more like a simple overlay.
• − Less sophisticated typography for the main title.

GPT Image 2

• + Excellent text rendering with stylzed 3D effects.
• + High-quality textures on the fish and rice that give a realistic PBR feel.
• + Rich composition with a well-defined diorama base.

• − Includes more decorative elements than the 'minimal garnish' requested.

Seedream 4.0

• + Adheres better to the 'minimal garnish' and simple plate request.
• + Clean cartoon aesthetic with soft lighting.
• + Accurate isometric perspective and solid background.

• − Text is somewhat flat and lacks the high-end 3D finish of the competition.
• − The sushi pieces look a bit more plastic and less 'refined' in texture.

GPT Image 2

• + Excellent anatomical proportions for all four animals.
• + Better lighting coherence and realistic interaction with the meadow floor.
• + Stunning detail on the fox and puppy fur textures.

• − The butterfly wings look a bit more synthetic compared to the animals.
• − Includes fewer 'dew sparkles' than the competing model.

Seedream 4.0

• + Captures the 'playfully chasing' and 'tumbling' dynamic more energetically.
• + High-quality dew sparkles and light bokeh throughout the scene.
• + Strong golden sunrise atmosphere with vibrant colors.

• − The fox's anatomy is distorted, especially the junction of the head and paws.
• − The kitten's pose and scale feel slightly supernatural and floaty.
• − The bunny has a slightly blurry face compared to the other subjects.

GPT Image 2

• + Excellent typography with beautiful serif details and correct accents.
• + Sophisticated composition with high-quality vector-style frame and shading.
• + Perfect adherence to the 'vintage' and 'textured paper' prompt elements.

• − The design is more complex than a strictly 'minimalist' logo usually allows.

Seedream 4.0

• + Simpler, more minimalist composition that feels modern-retro.
• + Good color balance between the deep brown and cream tones.
• + Accurate text rendering and placement on a banner.

• − The typography is less refined and lacks the classic elegance of the era.
• − The cloche shading feels a bit more like a generic illustration than a professional logo emblem.

GPT Image 2

• + Excellent text rendering with near-perfect spelling and typography.
• + Strict adherence to the 6-step chronological structure requested in the prompt.
• + Professional infographic layout with high-quality vector-style illustrations and consistent iconography.

• − The scale of the spacecraft in orbit is realistically too small for a vector icon style.
• − The lunar module illustration is slightly more detailed than a 'flat' vector style usually implies.

Seedream 4.0

• + Successfully captures a flat, simplified vector aesthetic for the Earth and Moon icons.
• + Adheres to the color palette well with a clean, light gray background.

• − Spelling errors in text labels such as 'Surfcce' and 'Tranquility' (though the latter is a common variant).
• − Cluttered and confusing composition that fails to cleanly separate the steps.
• − Inconsistent layout where 'Landing' is disconnected from the numbered sequence.